According to some experts in Computer Graphic's field these are three good metrics

- Physical realism;

- Photo realism;

- Functional realism.

Dear Stephan,

we use two concepts to define it, both are perceptual and usually more than enough to get the message through via computer generated illustrations or anymations.

1. Level of detail. In our medical illustrations done for Springer Medizin (see link to their outlet at the Science Photo Library), the level of detail can be compared to the map view of Google Map's Navigation. If you look at an entire body / city it wouldn't make sense to show every detail. It is enough to show only main arteries and veins (body) or main roads (navi). You wouldn't recognize anything if those main arteries (lit. for the body and fig. for streets) would be cluttered by small vessels down to the smallest arterioles and capillaries at such a global level.

In Google Maps, when you want to see more detail, you enlarge the view and Google maps adds more and more detail depending on the scale. In the medical illustration comparison this would be showing the whole body, part of the body in an macro anatomy level, then micro anatomy, cell level, detailed cell level, and molecular level.

2. Level of realism. Especially in scientific computer generated images, we use illustrative renderings to clarify things. Too much detail may distract the viewer. Even if an image has to be like a photograph, we usually don't show them in a real world light setting, but use a studio light setup instead where we can direct light in a way that enhances certain shapes, edges, curves or areas of interest. We may enhance colors and contrast.

In the medical image example, a realistic image would be mostly beige. Very diffiult to recognize something especially for non-medical people. We use certain color codes, e.g. red for blood vessels with fresh blood (mostly erteries) and blue for vessels with used blood (mostly veins). In the real world, blood vessels are wizen, twisting and winding. In anatomical illustration you usual show clean and round vessels.

Your pixel by pixel comparison ... well nobody will pay for such an extreme level of detail. In my experience, the highest level of realism is achieved in the movie industry, where CGI may be chaper than expensive real movie sets. Here you find the concept of "believability". The scene must be realistic enough to be believeable for the viewer, i.e. the viewer won't notice being cheated.

best regards

Stefan Sonntag

CAT PRODUCTION, Munich, Germany

http://www.sciencephoto.com/collection/springer-medicine

There is a concept of probabilistic inference. Consists of making inferences about certain properties of the scene based on the information in the image. I think this is a way to solve the problem. There are studies on the subject, I attached one.

Sorry for the late answer, but I was abroad.

The problem is "what is realism"? There's a big difference for instance between a "realistic" studio photo and a "candid" snapshot. The lighting, the colors are could be very different, yet both photos are considered realistic. I think "realism" depends on the subject and on the audience/target, as in photography.

Usually the "this-look-realistic" way could be measured by details (texture maps' accuracy, ratios among different subjects on the image and lighting). I found Mr. Interian's attach very interesting.

Yet since I'm just a CG amateur, I suggest you to put your question on a CG artists forum like RDNA (www.runtimedna.com), they will probably help you better than me.

Realism is a subjective standard. As an example, it's fairly common to make unreal adjustments to elements in an animation to make them appear more real. Furthermore, unreal adjustments or simple exaggerations may be made in an animation to better convey whatever idea is intended. 

I like that paper you attached, Ruben. I can see citing that in the future. Thanks.